Pooled log transformed odds ratios and 95% confidence intervals were calculated using an inverse variance random-effects model. Heterogeneity was assessed using the X2 and I2 statistics, with significance indicated when p<0.10 or I2>50%.
The influence of the following three potential causes of heterogeneity was examined using subgroup analyses and tests of interaction: severity of pneumonia; duration of proton pump inhibitor therapy; and proton pump inhibitor dose. These factors were determined a priori.
Two a priori sensitivity analysis were performed by excluding studies that defined pneumonia using only database coding and by excluding studies that did not adequately adjust for potential confounders (age, gender, smoking status, chronic obstructive pulmonary disease, previous stroke, functional status and gastro-oesophageal reflux disease). Post hoc sensitivity analysis was conducted by examining the influence of each of two studies with small degree of patient overlap.
Funnel plots were to be used to assess publication bias if more than 10 studies were included in the main analysis.